Can AI be less power hungry?

As an act of irony I thought I'd ask Chat GPT if AI could be less power hungry; This is what it said:

Yes — AI can become considerably less power-hungry, and there are several ways to do it.

Smaller models: Not every task needs a gigantic model. A small, specialised model can use a fraction of the electricity of a frontier model.

More efficient chips: AI accelerators are becoming much more efficient at the calculations involved.

Better algorithms: Techniques such as quantisation, sparsity and distillation allow models to do similar work with fewer calculations.

Use AI selectively: A simple question shouldn't require the computational equivalent of a huge research task. Systems can route easy requests to small models and difficult ones to larger models.

Reuse rather than recalculate: Caching common answers and computations can avoid doing the same work repeatedly.

Cleaner electricity: Even where energy use remains substantial, powering data centres with genuinely low-carbon electricity reduces the climate impact.

Smarter cooling: Data-centre cooling can consume a significant amount of energy and water, so better cooling systems matter too.

Longer-lived hardware: Manufacturing the chips and servers also has an environmental cost. Getting more useful work from existing equipment is important.

There's an interesting distinction, though: making AI more efficient doesn't necessarily mean total AI energy consumption will fall. If AI becomes ten times cheaper to run, people may simply use ten times as much of it — the classic rebound effect.

So the really important question isn't just "Can we make AI use less electricity per query?" It is "Can we make AI use less energy overall while still getting the benefits we actually need?"

And the answer to that is also yes — but it requires choosing efficiency over simply making ever-larger models.